Re: [SSI-devel] [PATCH] make migrate independent of /proc/<pid>/p in

[Aneesh K. K.V <ane...@di...>]

On Tue, 2003-08-12 at 00:31, Byrne, John (OpenSSI clusters - HP) wrote:
> Aneesh Kumar K.V wrote:
> > Hi,
> > 
> > I guess migrate command should work even if the pid is pinned to any
> > node. /proc/<pid>/pin should only be controlling the automatic load
> > balancing. The below patch fix this behaviour.( Tested on Alpha ) 
> > 
> > -aneesh 
> 
> You need to convince Bruce that this is the behavior he wants, first. 
> The current behavior of pin meets my understanding of what it is 
> supposed to do. I'd suspect that he'd prefer an option to migrate that 
> allowed you to override the pin setting, but we'll see.



Ok here is the way i looked at it. 

1) /proc/<pid>/pin can be set/unset by the user. 
2) user write 1 to pin the <pid> so that unexpected migration doesn't
happen. ( automatic load balancing. )
3) But user has the freedom to unset it and then migrate. ie,
	echo 0 > /proc/<pid>/pin 
	migrate <nodenum> <pid>

Since  above is already allowed why would one need to check pvp_pin when
using with migrate command. ( that's is executed  by the user ). I would
say /proc/<pid>/pin is to indicate <pid> don't  like unexpected
migration .  But if the user explicitly like to migrate then it should
be allowed. 


Returning migrate error as the write error seems to be a good idea. That
would also help the user to find out if the migrate has really happened
or not. 

-aneesh